[fpc-devel] NetBSD and OpenBSD...

Pierre Free Pascal pierre at freepascal.org
Thu May 31 16:47:41 CEST 2012


  Hello Liyuan,

  That really nice to hear,
I do commit some minor changes to those
targets, but I only have virtual machines
on which these systems run remotely.

  What would be nice is to have people testing 
those targets more deeply.
  In particular, it would be nice to get those two targets
as close as possible to their FreeBSD counterparts.

  After, it mainly depends on your interests,
if you want to debug OS specific problems,
a good sdtart it to look at the results of nightly testsuite runs.

  I run nightly testsuites for i386-netbsd target and i386-openbsd:
See, for netbsd:
http://www.freepascal.org/testsuite/cgi-bin/testsuite.cgi?os=6&cpu=1&version
=0&date=
and for openbsd:
http://www.freepascal.org/testsuite/cgi-bin/testsuite.cgi?os=7&cpu=1&version
=0&date=

And for reference, the corresponding freebsd results:


  The differences are probably mainly due to:

1) Errors in the OS specific part of the RTL,
(like for instance wrong values of some constants
in RTL sources)
2) errors in the compiler specific to the given OS (differences in ABI for
instance).
3) Packages that have not been tested for the OS.

The testsuite are run several times with different Comment fields in the web
page,
those correspond to different TEST_OPT variable passed to gmake 
when running the testsuite which alter slightly the results.

  Ideally we should have no difference in testsuite output,
no matter what the CPU or the OS is.
  Some tests are just failing for all CPU-OS pairs
as they correspond to things that are still not implemented in the compiler
(one that comes to my mind is support for sets of more than 256 elements,
which is not supported and the reason of the failure
test/tset6.pp).

  Another point of interest might be to know if the compiler
works on older release versions of the OS, the tests
are run of fairly recently installed VM's and I have no clue
about a minimal OS version...


  So you see that the ideas for help are numerous
and you can just try out and report your findings.

  All the above is about current trunk source, 
I didn't mention fixes branch, where comparison would also
be of interest...

  Anyhow, feel free to work on any topic and report 
about errors that you think you understood.
  The best here is usually to send a patch that allows to remove the
investigated error!

  Thanks for your interest in Free Pascal,

Pierre Muller




More information about the fpc-devel mailing list